import { defineConfig, devices } from '@playwright/test'; import { fileURLToPath } from 'node:url'; import { dirname, resolve } from 'node:path'; import { mkdirSync, writeFileSync } from 'node:fs'; import { execFileSync } from 'node:child_process'; import { join } from 'node:path'; import { createOwnedRunDir } from '../scripts/e2e-run-dir.mjs'; // __dirname under ESM. const __dirname = dirname(fileURLToPath(import.meta.url)); // Repo root is one level up from ui/. const repoRoot = resolve(__dirname, '..'); // Dedicated E2E port — avoids clashing with a dev server on 9876. if (!process.env.E2E_PORT) process.env.E2E_PORT = execFileSync(process.execPath, ['-e', "const s=require('net').createServer();s.listen(0,'127.0.0.1',()=>{console.log(s.address().port);s.close()})"], { encoding: 'utf8' }).trim(); const PORT = Number(process.env.E2E_PORT); const BASE_URL = `http://127.0.0.1:${PORT}`; // Isolated, throwaway data + workspace dirs so the E2E never touches real data // (real DB lives at ./data/maestro.db). A fresh temp DB also means the personal // space is auto-created clean on first /api/local/spaces hit. // The temp dir MUST be stable across the multiple times Playwright evaluates // this config (main process + worker), otherwise mkdtempSync would mint a new // dir per evaluation and specs could open a different DB from the webServer // (→ "no such table: jobs"). Self-seed it into the // environment on the FIRST evaluation: subsequent evals in this process reuse // it, and the webServer subprocess + worker processes inherit it via env. No // caller setup required (plain `npm run test:e2e` works); an explicitly-set // E2E_TMP is still honored. // E2E_RUN_OWNED が立っているのに実行用ディレクトリが無い状態は、手で // 環境変数を export したときにしか作れない。ここで先に落とす: 一時 // ディレクトリを 1 つでも作った後に投げると、config が完走せず // globalTeardown も走らないため、作った分が掃除されず残る。 if (process.env.E2E_RUN_OWNED === '1' && !process.env.E2E_RUN_TMP) { throw new Error('E2E_RUN_OWNED=1 but E2E_RUN_TMP is not set'); } const cache = join(repoRoot, 'node_modules', '.cache'); mkdirSync(cache, { recursive: true }); if (!process.env.E2E_TMP) { const ownedData = createOwnedRunDir(cache, 'maestro-e2e-data-'); process.env.E2E_TMP = ownedData.directory; process.env.E2E_DATA_TOKEN = ownedData.token; } const e2eTmp = process.env.E2E_TMP; if (process.env.E2E_RUN_OWNED !== '1') { const ownedRun = createOwnedRunDir(cache, 'maestro-e2e-run-'); process.env.E2E_RUN_TMP = ownedRun.directory; process.env.E2E_RUN_TOKEN = ownedRun.token; process.env.E2E_RUN_OWNED = '1'; } process.env.E2E_RUN_PREFIX = 'maestro-e2e-run-'; // 冒頭のガードが「OWNED=1 なのに未設定」を弾き、上の分岐が「未 OWNED」を // 必ず設定する。したがってここでは必ず値がある(TS はこの 2 段を追えない)。 const e2eRunTmp = process.env.E2E_RUN_TMP!; process.env.E2E_CACHE_ROOT = cache; const DB_PATH = join(e2eTmp, 'e2e.db'); const WORKTREE_DIR = join(e2eTmp, 'workspaces'); // Expose the throwaway DB path to specs that need to seed engine state which // can't be produced without a live LLM (e.g. a job parked for tool approval). // Written to a gitignored file next to this config so specs can read it. // Dedicated, deterministic config for the E2E webServer, handed over via // AAO_CONFIG. Two reasons: // 1) Isolation — without this, dist/main.js loads the repo-root ./config.yaml // if a developer happens to have one (it's gitignored, so present locally // but absent in CI). That silently changes E2E behavior between machines // (e.g. pointing the worker at a real LLM). Pinning AAO_CONFIG makes the // run identical everywhere; everything unset here falls back to defaults. // 2) SSH enabled — the settings sub-tab specs assert the space-scoped SSH // panel (space-ssh-panel) renders. That testid only mounts once // /api/ssh/connections answers 200; with SSH off the route 404s and the // panel shows its "SSH disabled" view (no testid). In isolation the spec // scraped the brief loading-state render and passed by luck; under full- // suite load that window closed and both SSH specs failed. Enabling SSH // (needs the MCP_ENCRYPTION_KEY set below) makes the panel deterministic. // 3) Bash outbound network allowed — WITHOUT this the webServer refuses to boot // on any host where bwrap cannot create a user namespace. safety.bash_sandbox // defaults to `auto`, and worker-bootstrap.ts (#1235) treats "network blocking // is configured but cannot take effect" as a startup error rather than a // silent downgrade. The dev container hits exactly that (the bwrap binary is // present, `--unshare-all` is denied), and whether a CI runner does depends on // its container settings — pinning this makes the boot outcome the same // everywhere instead of a property of the host. Nothing is actually opened up: // these specs never reach an LLM, so no agent Bash command ever runs. const E2E_CONFIG = join(e2eTmp, 'e2e-config.yaml'); writeFileSync(E2E_CONFIG, 'ssh:\n enabled: true\nsafety:\n bash_allow_network: true\n'); export default defineConfig({ testDir: resolve(__dirname, 'e2e'), // The build+boot can be slow; keep a roomy per-test budget. timeout: 60_000, expect: { timeout: 15_000 }, fullyParallel: false, workers: 1, retries: 0, reporter: [['list']], globalTeardown: resolve(__dirname, 'e2e-global-teardown.ts'), use: { baseURL: BASE_URL, headless: true, locale: 'ja-JP', // This sandbox disables the Chromium sandbox (bwrap/userns off). Launch // with --no-sandbox or every browser launch fails to spawn. launchOptions: { args: ['--no-sandbox', '--disable-setuid-sandbox', '--disable-dev-shm-usage'], }, trace: 'retain-on-failure', }, projects: [ { name: 'chromium', use: { ...devices['Desktop Chrome'] }, }, ], // Build before booting so a plain `npm run test:e2e` never serves a stale or // half-cleaned ui/dist tree. Previously the server could observe ui/dist/ // after Vite emptied it but before index.html was emitted, causing sendFile // ENOENT failures when E2E and a build ran near each other. // Boot the real orchestrator (worker mode) with auth OFF (the pinned // E2E_CONFIG has no auth section => defaults => synthetic 'local' user). // The webServer command builds both dist trees before starting Node. webServer: { command: 'node scripts/e2e-web-server.mjs', cwd: repoRoot, url: `${BASE_URL}/ui`, reuseExistingServer: false, timeout: 120_000, stdout: 'pipe', stderr: 'pipe', env: { ...process.env, PORT: String(PORT), E2E_RUN_TMP: e2eRunTmp, DB_PATH, WORKTREE_DIR, // Pin the config to the throwaway E2E file so the run never inherits a // developer's repo-root ./config.yaml (see E2E_CONFIG above). AAO_CONFIG: E2E_CONFIG, AAO_MODE: 'worker', LOG_LEVEL: 'warn', // Mount the MCP subsystem (gated on a 64-hex MCP_ENCRYPTION_KEY) so the // per-space MCP isolation E2E can register/list servers via /api/mcp/*. // A fixed throwaway key keeps the run deterministic; the temp DB it // encrypts is discarded after the run. MCP_ENCRYPTION_KEY: '00112233445566778899aabbccddeeff00112233445566778899aabbccddeeff', // Configure a STUB LLM so the first-run SetupWizard never gates the UI // (it shows when no model is configured). The endpoint is never actually // reached by these specs — they assert UI/DB state, not model output — so // a fake host is fine and keeps the E2E self-contained (no caller env). OLLAMA_BASE_URL: process.env.OLLAMA_BASE_URL ?? 'http://127.0.0.1:11434/v1', OLLAMA_MODEL: process.env.OLLAMA_MODEL ?? 'e2e-stub-model', }, }, });